OpenCode Creator: Why Open-Source AI Models Dominate. An interview with Dax Raad, co-founder of OpenCode, on why open-source AI coding tools are winning. Covers the business model, community dynamics, and the architectural decisions that make open-source agentic coding viable. MORE

Squidbleed (CVE-2026-47729). A Heartbleed-style heap buffer overread that has lived in every version of Squid Proxy since 1997, in its default configuration. Claude Mythos Preview found it by following an FTP directory-listing parser. A one-line check for the null terminator before strchr would have prevented it, but any attacker who controls an FTP server reachable by the proxy can leak other users' HTTP requests from memory. MORE

Measuring LLMs' Impact on N-day Exploits. Anthropic's Frontier Red Team tested models on 18 Firefox patches and 21 Windows kernel patches. Mythos Preview produced 8 working Firefox code-execution exploits and 8 full Windows privilege-escalation chains, autonomously. The first Windows PoC came in 31 minutes. N-day is now N-hour. The binding constraint to weaponizing a patch is now a few thousand dollars and API access. MORE

Claude's Criminally Bad Electron Mac App Is an Inside Job. John Gruber traces the answer to a single person: Felix Rieseberg, Anthropic's engineering lead for Claude desktop apps and one of the three members of the Electron project's Administrative Working Group. Rieseberg previously led the desktop teams for Slack and Notion, both famously non-native Electron experiences. The app isn't lazy engineering. It's a founding maintainer building on his own framework. MORE

Rewriting Bun in Rust. Bun, now owned by Anthropic, was 535,496 lines of Zig. Jarred Sumner used a pre-release Claude Fable 5 to port it to Rust in a single shot, line-for-line, with the same architecture and same test suite. A rewrite of this size would normally take a small team a full year. The result ships as Bun v1.4, eliminating entire classes of use-after-free and double-free bugs at the compiler level. MORE | RESPONSE

Introducing GPT-Live. OpenAI's new full-duplex voice model listens and speaks simultaneously, handling interruptions, pauses, and back-and-forth naturally. It delegates complex questions to GPT-5.5 in the background while maintaining the conversation. GPT-Live-1 substantially outperforms Advanced Voice Mode on GPQA scientific reasoning and BrowseComp web search tasks. MORE

The End of Reading Is Here. The Atlantic's cover story argues that universal literacy is no longer a reliable assumption. Students are arriving at college unable to read full books, attention spans have collapsed, and the age of deep reading is giving way to a post-literate culture shaped by screens and short-form content. MORE

Re:CACHE: 0-Click SXSS on Next.js. A two-stage cache-poisoning chain exploits reflected request headers to override the Content-Type of React Server Component payloads, turning them from text/x-component into text/html with injected scripts. Combined with a Refresh-header redirect on the homepage, it achieves zero-click stored XSS on the latest Next.js versions. Rewarded with a five-figure bounty. MORE

Europe's New Climate in Seven Charts. Two record-breaking heatwaves hit the UK and Europe in May and June, with the UK's June temperature record shattered by over 2C at some stations. Tropical nights, where temperatures stay above 20C, are spreading across England and Wales. Scientists say the warming trend will continue until net-zero emissions are reached. MORE

What's New in Biology: July 2026. A synthetic cell that eats, grows, and divides using only components scientists put inside it. The first base-edited human embryo, with single-letter DNA changes and none of the chromosomal damage CRISPR-Cas9 causes. Lab-grown human sperm and early egg cells. The first CRISPR medicine expands to children as young as two. MORE

T3MP3ST: Autonomous Red Teaming Platform. An open-source, multi-agent offensive-security framework from elder-plinius that turns your AI coding agent into a zero-day hunter. Keyless by design, it connects to Claude Code, Codex, or Hermes without API keys. Scores 90.1% pass@1 on XBOW's 104-challenge suite and found 8 of 10 held-out post-cutoff CVEs at exact file/line/CWE. All claims re-derive from committed data. MORE

XSRFProbe v3.0: A Ground-Up Rewrite. After six years of silence, the CSRF audit toolkit returns with a self-calibrating response-diffing engine that benchmarks each endpoint before testing it. Runs 26 distinct checks across token tampering, method override, origin validation, referer bypass, and SameSite exploitation. Every finding now carries severity and exploitability preconditions. MORE

@jinonehk - Security researcher

@J0_mart - Sharing is knowledge

@chevyphillip - Application Security Engineer and AI engineer pursuing a Master's in Data Science

@codebeast - Jack of a few trades. JavaScript, Teacher, Musician.

@jackhcable - Ethical hacker. CEO & Co-founder @ Corridor

Lost and Found. Hundreds of venues use one software tool to photograph and log lost items. Someone scraped the archives: thousands of accidental still-life portraits of stuffed animals, keys, sunglasses, and jackets, each tagged with place and date. A quiet museum of things people let go of. MORE

Kyoto in Sound. A composer-led walk through Kyoto's shrines, temples, and backstreets where you're handed a field recorder and binaural microphones instead of a camera. Afterwards, your recordings are mixed into a finished audio postcard. Close your eyes six months later and you're back on the path. MORE

Sci-Fi Corridor Archive. A beautifully curated collection of corridor scenes from sci-fi films, from Appleseed to X-Men to Planet of the Apes. An oddly satisfying exercise in pattern recognition across decades of production design. MORE

The Best Thing About The Satanic Panic. David Friedman argues that the phrase survived because it rhymes, exploring the rhyme-as-reason effect: people are more likely to believe something is true if it rhymes. The post includes a delightfully unhinged reduplicative poem covering everything from Zsa Zsa Gabor to Mahi Mahi. MORE

Small Penis Rule. According to Wikipedia, this is an informal strategy authors use to evade libel lawsuits: give a character based on a real person a small penis, and no one will come forward to claim the character is them. A 2021 law review article argues it's ineffective as a legal defense but effective as a deterrent. MORE

LifeOS: The AI Harness for Intent Engineering. An open-source platform that captures your goals, context, and definition of done, then conveys that intent to your AI on every task. Shift from prompt engineering to intent engineering: name where you are, name where you want to be, and the system climbs the gap. Works with Claude Code, Cursor, and Codex. MORE

Small Models Can Solve Big Problems. Jon Udell shows how Anthropic's Haiku categorizes community calendar events for less than a penny a day, solving an intractable classification problem that regex and word lists couldn't handle. The key insight: procedural code can't do it, but a low-powered LLM can, and curators get final say through an override mechanism. MORE

Inkdrop v6: Completed. Solo developer Takuya Matsuyama shipped every item on the 1.5-year roadmap. The note-taking app migrated from CodeMirror 5 to 6, LevelDB to SQLite, Webpack to electron-vite, and Flow to TypeScript. AI integrations now include inline proofreading, summarization, Mermaid diagrams, and next-edit suggestions that draw context from linked notes. MORE

Good Tools Are Invisible. Ginger Bill argues that a good tool disappears into the background. The telltale sign of a tool becoming identity is when users reframe limitations as features and sell the workaround as the reward. Productivity isn't how clever you felt, it's wall-clock time and mistakes made. The best tool is the one you forget you're using. MORE

18 Words: Daily Word Challenge. A daily word puzzle where you find a word using all the given letters before time runs out. Simple, clean, and the kind of small game that deserves a tab in your morning routine. MORE

AI Can't Recreate Thrust (But It Can Help You Understand It). James Randall asked Claude Code to recreate the classic 1986 BBC Micro game Thrust. It produced slop. But when he fed it the 6502 disassembly and asked questions, Claude became an extraordinary tool for software archaeology, extracting the exact physics timings and SN76489 sound chip behavior needed to build a pixel-perfect emulation. MORE

How a Writer Uses AI Without Losing His Voice. Craig Mod used to pay Campaign Monitor roughly $7,000 a year to send his newsletters. After rebuilding his own system with AI assistance, he cut that cost to near zero while keeping full control over his voice and cadence. A practical example of AI as tool, not replacement. MORE

The Correct Way to Build and Manage AI Agents. Jared Zoneraich, Builder in Residence at Cognition, explains why teams often over-engineer agent systems. The conversation covers when to use frameworks versus raw prompts, how to structure agent evaluation, and why the simplest approach usually wins. MORE

How I Run Autonomous Coding Agents from My Phone. Alessio Fanelli demonstrates two workflows using OpenAI Symphony and Linear to kick off coding agents from his phone, review results, and merge PRs without ever opening a laptop. The mobile IDE is real and it works. MORE

What Workflow Should You Get Your AI Agent to Do? (3 Questions). A framework for deciding what to automate: Can you specify success? Can you verify the output? Is the loop time fast enough? Three simple gates that separate good agent workflows from time sinks. MORE

The Bottleneck Might Be the Air in the Room. Mike Bowler brings a portable CO2 monitor to meetings and has watched it climb past 2,000 ppm. At 1,000 ppm, decision-making performance drops significantly on six of nine measures. At 2,500 ppm, some drop into the dysfunctional range. The cheapest variable in the building is the one nobody measures. Open a window. MORE

It Ain't Broke: Why Software Fundamentals Matter More Than Ever. Matt Pocock argues that AI coding tools are simultaneously overhyped and extraordinarily powerful. Used well, they're a force multiplier. Used without fundamentals, they generate junk at scale. Taste, judgment, and understanding of the underlying systems matter more, not less, as the tools get stronger. MORE

Growing on Purpose: The Work That Makes You. Jeremy Howard on human flourishing in the age of AI. A talk about finding work that develops you rather than depletes you, and why the question of what to do with AI is inseparable from the question of what to do with yourself. MORE

How to Actually Have an Elite Mindset. A conversation about what sits beneath burnout and exhaustion, and how to shift from surviving to performing at your best. Practical mental models for high performers who feel like they're running on empty. MORE

Legendary leg training session between bodybuilding icon Tom Platz and athlete David Hoffmann. The two reunite to focus on high-intensity squatting and leg development techniques, emphasizing that true growth requires pushing past physical comfort and embracing the "pain zone." MORE

Curated collection of books on infrastructure, maintenance, scaling, and engineering history.

Essay on how companies can lose their edge by rewarding comfort over careful engineering.

Manifesto arguing for AI that stays diverse, distributed, and human-centered.

Clear explainer on how neural networks, tokenization, embeddings, and transformers work.

An inside look at the architecture behind a modern agentic coding system.

and more….